The io context sharing introduced a per-ioc spinlock, that would protect
the cfq io context lookup. That is a regression from the original, since
we never needed any locking there because the ioc/cic were process private.

The cic lookup is changed from an rbtree construct to a radix tree, which
we can then use RCU to make the reader side lockless. That is the performance
critical path, modifying the radix tree is only done on process creation
(when that process first does IO, actually) and on process exit (if that
process has done IO).

As it so happens, radix trees are also much faster for this type of
lookup where the key is a pointer. It's a very sparse tree.
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

changes in the cfq for io_context sharing
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

Detach task state from ioc, instead keep track of how many processes
are accessing the ioc.
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

This is where it belongs and then it doesn't take up space for a
process that doesn't do IO.
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/cooloney/blackfin-2.6: (54 commits)
  MAINTIANERS: just use Mike gmail e-mail for contact and pawn the serial driver off onto Sonic
  [Blackfin] arch: remove old I2C BF54x porting.
  [Blackfin] arch: Add the semtimedop syscall.  Upstream uClibc doesn't compile without it.
  [Blackfin] arch: fix bug kernel boot message: memory information is not reasonable
  [Blackfin] arch: use common flash driver to setup partitions rather than the bf5xx-flash driver
  [Blackfin] arch: Fix bug - kernel build with Debug option enabled fails to boot up
  [Blackfin] arch: Fix bug Only RTC interrupt can wake up deeper sleep core.
  [Blackfin] arch: Add proper SW System Reset delay sequence
  [Blackfin] arch: Update copyright date
  [Blackfin] arch: GPIO API cleanup and anomaly update
  [Blackfin] arch: Fix BUG gpio_direction_output API is not compatitable with GENERIC_GPIO API interface
  [Blackfin] arch: Initial checkin of the memory protection support.
  [Blackfin] arch: set_bfin_dma_config shouldnt set SYNC or RESTART by default - add argument or option
  [Blackfin] arch: Add some comments - fix semicolons
  [Blackfin] arch: move all code related to CPLB handling into a new subdirectory under kernel/
  [Blackfin] arch: print out list of modules if kernel is crashing and tell people if the kernel is tainted 
  [Blackfin] arch: enable generic GPIO based I2C driver in STAMP-BF533, EZKIT-BF533 and EZKIT-BF561 boards
  [Blackfin] arch: Don't oops_in_progress if single step is comming from the kernel
  [Blackfin] arch: Fix BUG - kernel sometimes would stuck with KEYBOARD_GPIO on
  [Blackfin] arch: update to latest anomaly sheets
  ...

All the users of this driver explicitly specify the I2C bus numbers
to be used in their platform data. Make the driver respect that.
Signed-off-by: NKalle Pokki <kalle.pokki@eke.fi>
Cc: Bryan Wu <bryan.wu@analog.com>
Signed-off-by: NJean Delvare <khali@linux-fr.org>

Convert the i2c-au1550 bus driver to platform driver, and
register a platform device for the Alchemy Db/Pb series of
boards.
Signed-off-by: NManuel Lauss <mano@roarinelk.homelinux.net>
Signed-off-by: NJean Delvare <khali@linux-fr.org>

Zero-bytes transfers would leave the bus transaction unfinished
(no i2c stop is sent), with the following transfer actually
sending the slave address to the previously addressed device,
resulting in weird device failures (e.g. reset minute register
values in my RTC).
This patch instructs the controller to send an I2C STOP right after
the slave address in case of a zero-byte transfer.
Signed-off-by: NManuel Lauss <mano@roarinelk.homelinux.net>
Signed-off-by: NJean Delvare <khali@linux-fr.org>

Propagate the error values returned by i2c_wait() instead of overriding
them with a meaningless -1.
Signed-off-by: NJon Smirl <jonsmirl@gmail.com>
Signed-off-by: NJean Delvare <khali@linux-fr.org>

This adds a i2c_new_dummy() primitive to help work with devices
that consume multiple addresses, which include many I2C eeproms
and at least one RTC.
Signed-off-by: NDavid Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: NJean Delvare <khali@linux-fr.org>

It is no longer required to hold adapter->clist_lock to call
i2c_check_addr.
Signed-off-by: NJean Delvare <khali@linux-fr.org>
Cc: David Brownell <david-b@pacbell.net>

We only need to hold adapter->clist_lock when we touch the client list.
Signed-off-by: NJean Delvare <khali@linux-fr.org>

This goes on top of the patch removing most i2c_adapter.clients usage,
updating i2c_attach_client:

 - Don't call device_register() while holding clist_lock.  This
   removes a self-deadlock when on the i2c_driver.probe() path,
   for drivers that need to attach new devices (e.g. dummies).

 - Remove a redundant address check.  The driver model core does
   this as a consequence of guaranteeing unique names.

 - Move the "device registered" diagnostic so that it never lies;
   previously, on error paths it would falsely report success.
Signed-off-by: NDavid Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: NJean Delvare <khali@linux-fr.org>

The redundant i2c client list maintained by i2c-core is going away
soon, so drivers should stop using it now. Instead, they can use the
standard iterator provided by the device driver model
(device_for_each_child).
Signed-off-by: NJean Delvare <khali@linux-fr.org>
Cc: David Brownell <david-b@pacbell.net>
Cc: Michael Hunold <michael@mihu.de>

The i2c_adapter.clients list of i2c_client nodes duplicates driver
model state.  This patch starts removing that list, letting us remove
most existing users of those i2c-core lists.

 * The core I2C code now iterates over the driver model's list instead
   of the i2c-internal one in some places where it's safe:
      - Passing a command/ioctl to each client, a mechanims
        used almost exclusively by DVB adapters;
      - Device address checking, in both i2c-core and i2c-dev.

 * Provide i2c_verify_client() to use with driver model iterators.

 * Flag the relevant i2c_adapter and i2c_client fields as deprecated,
   to help prevent new users from appearing.

For the moment the list needs to stick around, since some issues show
up when deleting devices created by legacy I2C drivers.  (They don't
follow standard driver model rules.  Removing those devices can cause
self-deadlocks.)
Signed-off-by: NDavid Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: NJean Delvare <khali@linux-fr.org>

Add support for another variant of the VT8237. I couldn't test
I2C block support but I assume it is present as well.
Signed-off-by: NJean Delvare <khali@linux-fr.org>

Checking the PCI function number doesn't add any value, and it makes
adding dynamic IDs to the driver more difficult. Drop this check.
Signed-off-by: NJean Delvare <khali@linux-fr.org>

The i2c-amd756 and i2c-viapro drivers make use of the driver_data
field of the PCI device ID. When adding device IDs dynamically (by
writing to the new_id sysfs file) you cannot set the value of this
field by default. It has to be allowed explicitly. Do that, and
check the value so that the user can't crash the kernel accidentally.
Signed-off-by: NJean Delvare <khali@linux-fr.org>
Acked-by: NGreg Kroah-Hartman <gregkh@suse.de>

Add polling I2C transfer implementation for PXA I2C. This is needed
for cases where I2C transactions have to occur at times interrups are
disabled.
Signed-off-by: NMike Rapoport <mike@compulab.co.il>
Acked-by: Neric miao <eric.miao@marvell.com>
Signed-off-by: NJean Delvare <khali@linux-fr.org>

Allow i2c_transfer to be called in contexts where sleeping is not allowed.
It is the reponsability of the caller to ensure that the underlying i2c bus
driver will not sleep either.
Signed-off-by: NMike Rapoport <mike@compulab.co.il>
Signed-off-by: NJean Delvare <khali@linux-fr.org>

The implementation is unsafe, and anyway one can achieve the same from
userspace using i2c-dev + i2cdetect.

Also tag i2c_sibyte_add_bus __init.
Signed-off-by: NJean Delvare <khali@linux-fr.org>
Cc: Ralf Baechle <ralf@linux-mips.org>

Discard all I2C driver IDs that aren't used anywhere. That's not just a
couple of them, but more like 49 or one quarter of all defined IDs! And
this is just a first pass, next will come all IDs that are set but
never used, or used but never set.
Signed-off-by: NJean Delvare <khali@linux-fr.org>

I2C block read is supported since the ICH5. I couldn't get it to work
using the block buffer, so it's using the old-style byte-by-byte mode
for now.

Note: I'm also updating the driver author... The i2c-i801 driver was
really written by Mark Studebaker, even though he based his work on
the i2c-piix4 driver which was written by Philip Edelbrock.
Signed-off-by: NJean Delvare <khali@linux-fr.org>

Clear special mode bits (PEC, block buffer) at driver load time,
you never know in which state the device was left by its last user.

Also make sure that we reset the block buffer mode at the end of every
transaction, not only when PEC was used.
Signed-off-by: NJean Delvare <khali@linux-fr.org>

Use separate flags with explicit names to describe the features of
the ICH chip.
Signed-off-by: NJean Delvare <khali@linux-fr.org>

Provide a clearer documentation of which additional features each
ICH chip support, and which of these the driver supports.
Signed-off-by: NJean Delvare <khali@linux-fr.org>

Move the tps65010 header file from the OMAP arch directory to the
more generic <linux/i2c/...> directory, and remove the spurious
dependency of this driver on OMAP.
Signed-off-by: NDavid Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: NJean Delvare <khali@linux-fr.org>

Rename the main mutex in i2c-core from core_lists to core_lock. This
makes more sense now that the redundant lists are gone.
Signed-off-by: NJean Delvare <khali@linux-fr.org>

i2c_driver.list is superfluous, this list duplicates the one
maintained by the driver core. Drop it.
Signed-off-by: NJean Delvare <khali@linux-fr.org>
Acked-by: NDavid Brownell <dbrownell@users.sourceforge.net>

i2c_adapter.list is superfluous, this list duplicates the one
maintained by the driver core. Drop it.
Signed-off-by: NJean Delvare <khali@linux-fr.org>
Acked-by: NDavid Brownell <dbrownell@users.sourceforge.net>

Use more standard prototypes for i2c_use_client() and
i2c_release_client(). The former now returns a pointer to the client,
and the latter no longer returns anything. This matches what all other
subsystems do.
Signed-off-by: NJean Delvare <khali@linux-fr.org>
Cc: David Brownell <david-b@pacbell.net>

Don't implement our own reference counting mechanism for i2c clients
when the driver model already has one.
Signed-off-by: NJean Delvare <khali@linux-fr.org>
Cc: David Brownell <david-b@pacbell.net>

The framebuffer drivers for these pieces of hardware include support
for the DDC/I2C buses, so there is no need for separate drivers.
Signed-off-by: NJean Delvare <khali@linux-fr.org>

I2C adapter drivers are supposed to handle retries on nack by themselves
if they do, so there's no point in setting .retries if they don't.

As this retry mechanism is going away (at least in its current form),
clean this up now so that we don't get build failures later.
Signed-off-by: NJean Delvare <khali@linux-fr.org>

I2C driver IDs are optional, so if you don't need one, just omit it.
Signed-off-by: NJean Delvare <khali@linux-fr.org>

Signed-off-by: NJean Delvare <khali@linux-fr.org>

Signed-off-by: NJean Delvare <khali@linux-fr.org>
Acked-by: NAlessandro Zummo <alessandro.zummo@towertech.it>

This patch allows much of the I2C client address data to move from initdata
into text.
Signed-off-by: NMark M. Hoffman <mhoffman@lightlink.com>
Signed-off-by: NJean Delvare <khali@linux-fr.org>

The 10-bit address support in i2c-algo-pcf is so heavily broken that
it can't have ever been used. Nobody ever complained, so I'll take it
that nobody needs it. Let's just delete it.
Signed-off-by: NJean Delvare <khali@linux-fr.org>

Update comments and logging on return path for byte writes.  NAK is
an error, to be reported or optionally ignored.  Timeouts are always
errors.  Lost arbitration is not currently handled, so don't even list
it as an option in the error message.

Don't return bogus EFAULT code for inappropriate NAK; EIO is better,
there is no bad userspace address in question.
Signed-off-by: NDavid Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: NJean Delvare <khali@linux-fr.org>